Comparison of a neural network and a regression model to estimate suspended sediment in a semiarid basin

Susanne Schnabel, Marco Maneta

Research output: Contribution to journalArticlepeer-review

Abstract

In small semiarid basins with ephemeral flows most of the sediment conveyance takes place during runoff peaks exceeding a certain discharge. Sediment load is commonly modelled using rating curves fitting the water-sediment discharge relationship. The performance of a feed-forward back-propagation artificial neural network (ANN) and a multiple quadratic regression (MQR) model are tested using data from the Parapuños Catchment, a wooded rangeland located in SW Spain. Both models were calibrated using rainfall and discharge time series and derived variables such as rainfall intensity, runoff coefficient and rate of change of discharge. The final set of variables used in the analysis was done based on sensitivity analysis for the ANN model and based on an analysis of statistical significance of parameters in the MQR model. The performance of ANN and MQR were similar but better than rating curves of a single variable. In addition, ANN and MQR can reproduce the hysteretic loop of the sediment-discharge relationship.

Original languageEnglish
Pages (from-to)91-100
Number of pages10
JournalIAHS-AISH Publication
Issue number299
StatePublished - 2005

Keywords

  • Modelling
  • Neural network
  • Semiarid environments
  • Suspended sediment

Fingerprint

Dive into the research topics of 'Comparison of a neural network and a regression model to estimate suspended sediment in a semiarid basin'. Together they form a unique fingerprint.

Cite this